To: tjo@acm.org
Cc: pgsql-sql@postgresql.org
Subject: Re: 9.17.5. Row-wise Comparison 
In-reply-to: <4254AD87.2030508@acm.org> 
References: <200504032037.j33Kb1kh016358@hosting.commandprompt.com>
	<4254AD87.2030508@acm.org>
Comments: In-reply-to "TJ O'Donnell" <tjo@acm.org>
	message dated "Wed, 06 Apr 2005 20:48:23 -0700"
Date: Thu, 07 Apr 2005 00:47:42 -0400
Message-ID: <13655.1112849262@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

"TJ O'Donnell" <tjo@acm.org> writes:
> I've been using syntax like
> select a from tbl where (b,c,d) > (1,2,3)
>    to mean
> select a from t where b>1 and b>2 and d>3

> But I see in the manual at:
> http://www.postgresql.org/docs/7.4/interactive/functions-comparisons.html#AEN12735
> that only = and <> operators are supported.  Does this section of the manual
> not properly apply to this query?  Is the manual in error, or am I
> not understanding?

PG's current code acts as you are supposing, but it is broken because it
doesn't follow the SQL spec, and we will change it as soon as someone
gets around to working on it.

The spec says that this syntax implies a column-by-column ordering,
essentially

	if (b > 1) then true
	else if (b = 1 and c > 2) then true
	else if (b = 1 and c = 2 and d > 3) then true
	else false

You can find related discussions in the archives from a few months
back.  The spec's semantics correspond exactly to the sort ordering
of a multiple-column btree index, and so there are good reasons why we'd
want to provide that behavior even if it weren't mandated by the spec.

			regards, tom lane